Recently, OpenAI released their GPT-4 API to beta testers and ever since then, developers have been experimenting with implementing the model differently, so it can be used as an “agent” to complete tasks with as little human assistance as possible. Two open source projects have recently attracted attention: AutoGPT and BabyAGI.
AutoGPT, developed by Toran Bruce Richards, is an experimental open source application that demonstrates the capabilities of GPT-4. AutoGPT works by taking output from GPT-4 and then, adding an improvised external memory, it loops back into itself in order to improve a task, recognize mistakes, and offer suggestions on how to improve it. While the script is still far from being fully autonomic, in an ideal condition, it could help complete tasks without the help of a human. The script can be tested locally, as long as an OpenAI API key and a Google search API key are provided.
On the other hand, BabyAGI, developed by Yohei Nakajima takes on a task-oriented approach. According to Nakajima, he was inspired by the “HustleGPT” movement, which strives to construct businesses automatically with the help of GPT-4.
However, both projects still do not possess the level of autonomy and ability to adapt to new situations as expected. They still lack the capacity to buy products, rather they can search and look for a product that satisfies the criteria. Though, with the right purchasing API, it could become a reality. There have also been reports and cases of AI making up content, known as “confabulations”.
While these projects will still take years to progress the way they should, they serve as a proof that agents can be used effectively to complete tasks and could be more capable than a single AI model. The “AutoGPT” and “BabyAGI” projects have generated lots of hype and made developers interested in using GPT-4 for projects considered impossible in the past.
Toran Bruce Richards is the founder of AutoGPT and his ultimate goal is to create Artificial General Intelligence (AGI). In AI, “general intelligence” is the hypothetical capability of an AI system to be able to complete tasks without pre-defined rules or patterns, in contrast to “narrow AI” that only functions in a specific context.
Yohei Nakajima, the founder of BabyAGI, is striving to create what he dubs as “GodMode” — a web-based version of his script. It is apparent that there is still plenty of work to be done before autonomous AI agents develop a full range of AI tools and complete complex tasks efficiently by themselves.
